Some reflections on COLM and TADA 2026

Note: I originally wrote these reflections to share with my friends. I ended up getting super encouraging feedback and some even found my takeaways to be helpful while attending EMNLP this year, hence I decided to publish this piece here. A lot of the observations about the state of research in the AI era seem to have struck a nerve with others' experiences at other conferences ([1], [2]). For sake of anonymity, I have removed all researcher names; However, I'm always happy to spill the tea in person xD. For those interested, I presented "The Garden of Forking Prompts" at TADA and COLM. My poster is available here.

Slop is everywhere, and is fairly easy to spot

I found many talks/presentations slides as well as most of the posters to have clear Claude "tells". Now, I didn't immediately think this was bad (I used AI assistance to make my poster too!) but I found myself less interested in interacting with a poster the moment I spotted these cues. My takeaway was: If you are going to use AI to generate slides/posters, at least take the time to remove the obvious tells. I actually found myself highly engaged at a poster where there were some clearly misaligned columns (signalling that the author did not use AI). Ultimately, this is not a value judgement, people may be using AI for all sorts of things and they may actually have put effort into the poster—but as a presenter, it's a lesson that some very judgemental people (like me!) might be discouraged by obvious signs of heavy AI use.

Attending conferences is exhausting

This is probably obvious to people who have attended multiple conferences before, but as an early career academic trying to make the most out of conferences, I think it's worth reminding myself for the next one. There were so many events happening during and after each conference day. Especially in SF, I got accepted into multiple socials happening after conference. But at the end, I realized my time would be better spent skipping some of these events; even skipping some conference things. On my last day I took some time off from the workshops in the morning and in hindsight, taking the third or fourth morning would've resulted in better focus (and just more fun!). I don't have too many regrets here–since I visited SF earlier this year, but this is something I would definitely do more next time!

Being awkward is okay(?)

Going up to strangers and striking up conversations is hard. It's harder still when they are someone you really look up to. I had a bunch of these encounters. In hindsight, I think this is okay. Most researchers have been in the place that you are and they understand (right?). Some might even appreciate the "courage". I got rewarded a couple times and other times I sensed the other person wanted an out. My policy in such cases was to quickly thank them for their time and move on. With luck, they won't remember this interaction and at the next conference you can always say "we met at COLM last year". With more junior researchers, a great way to peel off is ask for their Bluesky/Linkedin. This worked like a charm.

The poster pitch should be tailored

Another one of the lessons I learnt while presenting the work was that you need to gauge people's interest before you start talking to them about your poster. A senior researcher asked me for a spiel and in a few minutes had to interrupt me and leave. This was a bit of a fumble on my part not knowing which version of my pitch was warranted in this situation. My last "spiel" of the day (with one of the legends in NLP) went much better. He asked for a 30 second version and had many follow-ups. Again, I think that the senior researcher will forgive me, but it was a good learning experience.

Managing the poster crowd

In my long list of fumbles ("Learning Experiences"), there was one at TADA where I was doing my spiel in batches and some people had to wait 5-10 minutes at the poster before I was done with questions from the last group. I did not notice that a researcher I admire a lot (but hadn't interacted in person before) waited for quite some time and I wasn't able to give them the best interaction. Now, this is going to happen, there will be researchers you don't know by face and hence might not be able to give special attention to—nor do I think separating your poster attendees into "famous and not famous people" is a good idea—but managing the crowd is part of your job. Once you do your spiel and an adequate number of questions has been answered, do not hesitate to intentionally talk to the people in the back. This will send a signal to the group in the front that you want to rotate the audience, but also avoids attrition. This varies case by case, but I think I was mostly afraid of seeming "rude", which on looking back, nobody cares about.

Go socialize!

This is a little bit at odds with a point from earlier ("Attending conferences.."), but is worth emphasizing. At all the conferences I've attended, best parts were talking to peers and researchers outside the conference. The dinner after TADA led to me making friends with one of the other PhD students and we ended up hanging out a bunch. I got boba tea with a PhD student I once chatted with on day 2 which was a great way to socialize and get away from the conference venue. The socials always led to me meeting up with the unlikeliest of people. Some of these will be trial-and-errors. Sometimes pushing your boundaries will just result in more fatigue, but eventually you'll find (hopefully) what works best for you. Especially at ACL events where we have Whova, going to sightsee with a few people you've never met before can lead to great friendships!

On notetaking

Conferences (especially NLP/ML) can be huge. I made it a point to sit down whenever I was by myself and make quick notes about: people I chatted with, talks I liked, ideas I had, etc. This is a great way of keeping track of the potential collaborations and connections in future. I also experimented with having a physical notebook and pen for this. Mostly because when I whip out my phone and start typing in the middle of a conference, I worry about offending the other person (signalling that I am uninterested or distracted). I highly recommend doing some version of this.